Experimental comparison of unsupervised methods for deinterleaving pulse trains

The aim of this article is to compare recent unsupervised methods for deinterleaving pulse trains in a radar context. The comparisons are made for two types of data: synthetic data generated with semi-Markov processes and realistic data generated with an electronic warfare simulator. We also propose a robustness analysis of these different algorithms in various perturbation scenarios, modeled with the addition of measurement noise or the introduction of missingness mechanisms.